ExtractFoxExtract code from a screenshot, video, or PDF
Drop a screenshot, video frame, slide, or PDF that contains code — get clean source back with indentation preserved, language detected, and ready to paste straight into your editor. Skips the OCR mess of Tesseract on monospaced fonts and the manual cleanup that follows.
What should we pull from this code source?
Or pick specific fields
Or describe it yourself
Why this matters
Generic OCR mangles code. Spaces and tabs blur, brackets land on the wrong line, single-character tokens (i, l, 1, |) get confused. ExtractFox uses a multimodal model that reads source code semantically — it knows what a function declaration looks like, where indentation matters, and what to do with a string that contains escape characters. Conference slides put 15 lines of Rust on a dark background and Tesseract reads 0 as O and collapses four-space indents into one — the pasted code doesn't compile and you spend 10 minutes fixing it. Paywalled tutorials and copy-protected docs block select-all, so the only option is a screenshot and a prayer that OCR gets the closing braces right.
How it works
- Step 1Upload the source
A screenshot, slide image, video frame, conference talk PDF, or a copy-protected web page screenshot. Multiple files at once.
- Step 2Pick a mode
Code only (clean), code with line numbers preserved, code with comments, or all code blocks from a multi-snippet image.
- Step 3Copy or download
Pick Clean code and the result is the code by itself — copy it straight into your editor or download it as .txt. The multi-block modes come back as a table (one row per block) you can take as Excel, CSV, or JSON.
Common use cases
Sample output
Example: a screenshot of a Python function from a tutorial
| language | python |
| code | def fibonacci(n): if n <= 1: return n a, b = 0, 1 for _ in range(n - 1): a, b = b, a + b return b # Print the first 10 Fibonacci numbers for i in range(10): print(fibonacci(i)) |
Frequently asked questions
How do I extract code from a screenshot?+
Upload the screenshot here, pick a mode (clean code, with line numbers, all blocks, etc.), and click Extract. Output is ready to paste into your editor with indentation preserved.
Does it detect the language?+
Yes — every mode returns a detected language so you can save the file with the right extension. Recognized languages include Python, JavaScript, TypeScript, Go, Rust, Java, C++, Ruby, PHP, SQL, shell, and dozens more.
Can I extract code from a video tutorial?+
Take a screenshot of the frame with the code, then upload it. For longer tutorials with code that scrolls, screenshot each state and use the All code blocks mode to consolidate.
What about REPL or Jupyter notebook screenshots?+
Pick the Code and its output mode. Each input cell and its corresponding output come back separately so you can re-run the code without the output mixed in.
Will indentation be exact?+
Yes. Tabs vs spaces is preserved, indentation depth is preserved. Code that runs in the original screenshot will run when you paste the extraction.
How is this different from generic OCR like Tesseract?+
Tesseract treats code as text and routinely loses indentation, confuses similar characters (l/1/i, O/0), and breaks on monospaced fonts with thin strokes. The multimodal model reads code semantically — it knows what valid syntax looks like and produces output that compiles.
Can it recover code from a photo of a whiteboard or projector screen?+
Yes. Whiteboard and projector photos with moderate glare and angle work — indentation and brackets are preserved. Very faint marker or extreme distance may need a closer photo.
Does it handle syntax-highlighted code with colored backgrounds?+
Yes. Dark themes, light themes, and rainbow syntax highlighting are all fine — the model reads token structure, not font color.